From the case that the article is presenting, I think LLMs are acting as a validation step – the average person who doesn't know how to code creates a minimal, LLM-spaghetti system (e.g. a restaurant menu website) with ChatGPT, validate if this is something that they need, iterate on it, create a specification, and then bring in an actual (costly) engineer that can fix and improve the system.

There's a lot of LLM byproducts that leave us in bad taste (hate all of the LLM slop on the internet), but I don't think this is one of them.